Hi James,
That is a detail that we probably should have noted in our paper. The decision to perturb humidity was arbitrary.
I think that it would be reasonable to perturb any or all of the boundary condition (BC) fields by some small amount. In our case, we used humidity because it was the first field that we decided to perturb when trying to demonstrate that sensitivity experiments can produce large and coherent precipitation anomalies that are nothing more than 'model noise'. The humidity perturbation experiments demonstrated this thoroughly, so we never went beyond perturbing humidity in our study.
I'm not sure I have great advice for how to choose which field(s) to perturb. Since we were trying to demonstrate that RegCM is quite sensitive to boundary conditions, it served well to perturb only one BC field by a tiny amount. However, we had no a priori reason for choosing humidity. In retrospect though, I wonder if it might be a better choice to perturb all the BC fields by a small amount if you're trying to sample the range of model states that can result from a given set of (uncertain) boundary conditions. Then the problem would be to decide how large the perturbation should be for each BC field--this likely depends on the field and on the estimated uncertainty in the boundary conditions being used for the simulation. As far as I'm aware, this isn't something that has been rigorously studied.
Cheers,
-Travis-
Travis O'Brien
Lawrence Berkeley National Lab
(510) 495-8047